The dataset is currently empty. Upload or create new data files. Then, you will be able to explore them in the Dataset Viewer.
How Reliable Are Models' Self-Logs?
When a model is asked to keep its own action log, is that log true? Each self-written entry is compared with an automatic record of the model's tool calls and labelled as accurate, softened, fabricated or omitted.
Status: planned, not yet released. The benchmark scenarios are still in review and no study runs have been made. This repository holds no data yet; it is a placeholder for the future release.
| Data | None yet |
| Code | GitHub |
| Earlier study | INSIDER_LLM_DETECTION_BENCHMARK (archived) |
- Downloads last month
- 12